Linux 5.15 I/O Can Achieve Up To ~3.5M IOPS Per-Core

[phoronix]

Phoronix: Linux 5.15 I/O Can Achieve Up To ~3.5M IOPS Per-Core

In addition to the block subsystem changes submitted for the Linux 5.15 merge window, Jens Axboe also sent in a separate pull request for this new kernel cycle to provide support for bio recycling. In turn this can enhance the Linux I/O limits by around 10%...

https://www.phoronix.com/scan.php?pa...5-IO-3.5M-IOPS

[perpetually high]

5.15 shaping up to be a banger. Hopefully an LTS kernel.

[labyrinth153]

Saw an article where Microsoft was doing 13.7 million ops on optane. https://techcommunity.microsoft.com/...ws/ba-p/428314

[MadeUpName]

Saw an article where Microsoft was doing 13.7 million ops on optane. https://techcommunity.microsoft.com/...ws/ba-p/428314

For the entire cluster not just per core.

With 100% reads, the cluster delivers 13,798,674 IOPS.

[F.Ultra]

For the entire cluster not just per core.

With 100% reads, the cluster delivers 13,798,674 IOPS.

Which was on a total of 672 cores :-)

[labyrinth153]

Thanks! Not to sound lazy, but I was curious of the specifics and didn’t have time to dig in.

[coder]

It'd be interesting if he also quoted IOPS/W, for the CPU. I presume it also went up, but it'd be nice to know.

[Sadhu]

He also got over 15M IOPS on his box with 3 optane drives: https://twitter.com/axboe/status/143...949981706?s=20

[markg85]

Thanks! Not to sound lazy, but I was curious of the specifics and didn’t have time to dig in.

A comment like that makes me mad.
It means you don't read, you don't care and just from a super quick google search think "look dorks, this is faster".
Then to just dump it here and let us figure out, you said so in your very post... That's downright insulting. It means you consider your time more valuable then ours.

I, respectfully, ask you to just don't do that!

[labyrinth153]

No offense (some offense) but cry about it. I apologized already and thats all you get, you don’t get to lecture me too.